Segmentation Metrics Research Articles

Deep learning kidney segmentation with very limited training data using a cascaded convolution neural network.

BackgroundDeep learning segmentation requires large datasets with ground truth. Image annotation is time consuming and leads to shortages of ground truth data for clinical imaging. This study is to investigate the feasibility of kidney segmentation using deep learning convolution neural network (CNN) models trained with MR images from only a few subjects.MethodsA total of 60 subjects from two cohorts were included in this study. The first cohort of 20 subjects from publicly available data was used for training and testing. The second cohort of 40 subjects with renal masses from our institution was used for testing only. A few-shot deep learning approach using 3D augmentation was investigated. T1-weighted images in the first cohort were used for training and testing. Cascaded CNN networks were trained using images from one, three, and six subjects, respectively. Images for the remaining subjects were used for testing. Images in the second cohort were utilized for testing only. Dice and Jaccard coefficients were generated to evaluate the performance of CNN models. Statistical analyses for segmentation metrics among different approaches were performed.ResultsOur approach achieved mean Dice coefficients of 0.85 using a single training subject and 0.91 with six training subjects. Compared to a single Unet, the cascaded network significantly improved the results using a single training subject (Dice, 0.759 vs. 0.835; p<0.001) and three subjects (0.864 vs. 0.893; p = 0.015) in the first cohort, and the results for the second cohort (0.821 vs. 0.873; p = 0.008).ConclusionOur few-shot kidney segmentation approach using 3D augmentation achieved a good performance even using a single Unet. Furthermore, the cascaded network significantly improved the performance of segmentation and was superior to a single Unet in certain cases. Our approach provides a promising solution to segmentation in medical imaging when the number of ground truth masks is limited.

DDTNet: A dense dual-task network for tumor-infiltrating lymphocyte detection and segmentation in histopathological images of breast cancer.

The morphological evaluation of tumor-infiltrating lymphocytes (TILs) in hematoxylin and eosin (H& E)-stained histopathological images is the key to breast cancer (BCa) diagnosis, prognosis, and therapeutic response prediction. For now, the qualitative assessment of TILs is carried out by pathologists, and computer-aided automatic lymphocyte measurement is still a great challenge because of the small size and complex distribution of lymphocytes. In this paper, we propose a novel dense dual-task network (DDTNet) to simultaneously achieve automatic TIL detection and segmentation in histopathological images. DDTNet consists of a backbone network (i.e., feature pyramid network) for extracting multi-scale morphological characteristics of TILs, a detection module for the localization of TIL centers, and a segmentation module for the delineation of TIL boundaries, where a boundary-aware branch is further used to provide a shape prior to segmentation. An effective feature fusion strategy is utilized to introduce multi-scale features with lymphocyte location information from highly correlated branches for precise segmentation. Experiments on three independent lymphocyte datasets of BCa demonstrate that DDTNet outperforms other advanced methods in detection and segmentation metrics. As part of this work, we also propose a semi-automatic method (TILAnno) to generate high-quality boundary annotations for TILs in H& E-stained histopathological images. TILAnno is used to produce a new lymphocyte dataset that contains 5029 annotated lymphocyte boundaries, which have been released to facilitate computational histopathology in the future.

Introduction of Lazy Luna an automatic software-driven multilevel comparison of ventricular function quantification in cardiovascular magnetic resonance imaging

Cardiovascular magnetic resonance imaging is the gold standard for cardiac function assessment. Quantification of clinical results (CR) requires precise segmentation. Clinicians statistically compare CRs to ensure reproducibility. Convolutional Neural Network developers compare their results via metrics. Aim: Introducing software capable of automatic multilevel comparison. A multilevel analysis covering segmentations and CRs builds on a generic software backend. Metrics and CRs are calculated with geometric accuracy. Segmentations and CRs are connected to track errors and their effects. An interactive GUI makes the software accessible to different users. The software’s multilevel comparison was tested on a use case based on cardiac function assessment. The software shows good reader agreement in CRs and segmentation metrics (Dice > 90%). Decomposing differences by cardiac position revealed excellent agreement in midventricular slices: > 90% but poorer segmentations in apical (> 71%) and basal slices (> 74%). Further decomposition by contour type locates the largest millilitre differences in the basal right cavity (> 3 ml). Visual inspection shows these differences being caused by different basal slice choices. The software illuminated reader differences on several levels. Producing spreadsheets and figures concerning metric values and CR differences was automated. A multilevel reader comparison is feasible and extendable to other cardiac structures in the future.

Decentralized Distributed Multi-institutional PET Image Segmentation Using a Federated Deep Learning Framework.

The generalizability and trustworthiness of deep learning (DL)-based algorithms depend on the size and heterogeneity of training datasets. However, because of patient privacy concerns and ethical and legal issues, sharing medical images between different centers is restricted. Our objective is to build a federated DL-based framework for PET image segmentation utilizing a multicentric dataset and to compare its performance with the centralized DL approach. PET images from 405 head and neck cancer patients from 9 different centers formed the basis of this study. All tumors were segmented manually. PET images converted to SUV maps were resampled to isotropic voxels (3 × 3 × 3 mm3) and then normalized. PET image subvolumes (12 × 12 × 12 cm3) consisting of whole tumors and background were analyzed. Data from each center were divided into train/validation (80% of patients) and test sets (20% of patients). The modified R2U-Net was used as core DL model. A parallel federated DL model was developed and compared with the centralized approach where the data sets are pooled to one server. Segmentation metrics, including Dice similarity and Jaccard coefficients, percent relative errors (RE%) of SUVpeak, SUVmean, SUVmedian, SUVmax, metabolic tumor volume, and total lesion glycolysis were computed and compared with manual delineations. The performance of the centralized versus federated DL methods was nearly identical for segmentation metrics: Dice (0.84 ± 0.06 vs 0.84 ± 0.05) and Jaccard (0.73 ± 0.08 vs 0.73 ± 0.07). For quantitative PET parameters, we obtained comparable RE% for SUVmean (6.43% ± 4.72% vs 6.61% ± 5.42%), metabolic tumor volume (12.2% ± 16.2% vs 12.1% ± 15.89%), and total lesion glycolysis (6.93% ± 9.6% vs 7.07% ± 9.85%) and negligible RE% for SUVmax and SUVpeak. No significant differences in performance (P > 0.05) between the 2 frameworks (centralized vs federated) were observed. The developed federated DL model achieved comparable quantitative performance with respect to the centralized DL model. Federated DL models could provide robust and generalizable segmentation, while addressing patient privacy and legal and ethical issues in clinical data sharing.

ARACAM: A RGB-D Multi-View Photogrammetry System for Lower Limb 3D Reconstruction Applications.

In the world, there is a growing need for lower limb prostheses due to a rising number of amputations caused primarily, by diabetic foot. Researchers enable functional and comfortable prostheses through prosthetic design by integrating new technologies applied to the traditional handcrafted method for prosthesis fabrication that is still current. That is why computer vision shows to be a promising tool for the integration of 3D reconstruction that may be useful for prosthetic design. This work has the objective to design, prototype, and test a functional system to scan plaster cast molds, which may serve as a platform for future technologies for lower limb reconstruction applications. The image capture system comprises 5 stereoscopic color and depth cameras, each with 4 DOF mountings on an enveloping frame, as well as algorithms for calibration, segmentation, registration, and surface reconstruction. The segmentation metrics of dice coefficient and Hausdorff distance (HD) show strong visual similarity with an average similarity of 87% and average error of 6.40 mm, respectively. Moving forward, the system was tested on a known 3D printed model obtained from a computer tomography scan to which comparison results via HD show an average error of ≤1.93 mm thereby making the system competitive against the systems reviewed from the state-of-the-art.

Using Conceptual Recurrence and Consistency Metrics for Topic Segmentation in Debate

We propose a topic segmentation model, CSseg (Conceptual Similarity-segmenter), for debates based on conceptual recurrence and debate consistency metrics. We research whether the conceptual similarity of conceptual recurrence and debate consistency metrics relate to topic segmentation. Conceptual similarity is a similarity between utterances in conceptual recurrence analysis, and debate consistency metrics represent the internal coherence properties that maintain the debate topic in interactions between participants. Based on the research question, CSseg segments transcripts by applying similarity cohesion methods based on conceptual similarities; the topic segmentation is affected by applying weights to conceptual similarities having debate internal consistency properties, including other-continuity, self-continuity, chains of arguments and counterarguments, and the topic guide of moderator. CSseg provides a user-driven topic segmentation by allowing the user to adjust the weights of the similarity cohesion methods and debate consistency metrics. It takes an approach that alleviates the problem whereby each person judges the topic segments differently in debates and multi-party discourse. We implemented the prototype of CSseg by utilizing the Korean TV debate program MBC 100-Minute Debate and analyzed the results by use cases. We compared CSseg and a previous model LCseg (Lexical Cohesion-segmenter) with the evaluation metrics Pk and WD. CSseg had greater performance than LCseg in debates.

Efficient blood vessel segmentation from color fundus image using deep neural network

Blood vessel segmentation is an essential element of automatic retinal disease screening systems. In particular, retinal blood vessel analysis from fundus image is vital in the identification and diagnosis of cardiovascular and ophthalmological diseases (Ex: Diabetic Retinopathy, Macular degeneration, Retinal Pigmentosa, Macular Edema, and various stages of Glaucoma, etc). Wherefore, the diagnosis of these diseases by automatic vessel segmentation has become essential, especially in disclosure of premature prognosis of vision condition. In general, blood vessel extraction is divided into vessel tracking and pixel classification. In vessel tracking a vasculature model is expanded from a seed point. In pixel classification, the classifier classifies the pixels as either a vessel or background pixel, which is demonstrated in the proposed architecture. In this paper, deep learning based 19 layer U-Net architecture is proposed for the accurate and efficient segmentation of blood vessels. Prior to segmentation, a pre-processing block of AlexNet architecture is introduced for the classification of high-quality images from the experimented databases. This pre-classification stage helps in efficiently picking high-quality images determined by clarity, field definition, and sharpness. AlexNet classification is pivotal in enhancing the overall performance of the system by segmenting fine and tiny blood vessels. The proposed U-Net architecture has an encoder-decoder framework with 9 and 5 convolutional layers in each respectively. In order to boost the efficiency of the network as well as to reduce training and testing time, a proper choice of kernel dimension and number of filters are necessary. Our architecture was investigated on popular databases such as DRIVE, ARIA_d and MESSIDOR and various performance measures (accuracy, sensitivity, specificity, sensibility, Dice coefficient, and Jaccard coefficient) have been computed along with the Receiver Operating Characteristics. It is observed that the accuracy for DRIVE, ARIA_d and MESSIDOR are 90.60%, 87.60% and 83.42%, respectively. Area under curve in Receiver Operating Characteristics plot is found to be 98.54%, 93.28% and 88.18%, for DRIVE, ARIA_d and MESSIDOR databases, respectively. Results with the proposed architecture show remarkable improvement in the performance metrics for blood vessel segmentation.

Exploratory Analysis on Pixelwise Image Segmentation Metrics with an Application in Proximal Sensing

A considerable number of metrics can be used to evaluate the performance of machine learning algorithms. While much work is dedicated to the study and improvement of data quality and models’ performance, much less research is focused on the study of these evaluation metrics, their intrinsic relationship, the interplay of the influence among the metrics, the models, the data, and the environments and conditions in which they are to be applied. While some works have been conducted on general machine learning tasks such as classification, fewer efforts have been dedicated to more complex problems such as object detection and image segmentation, in which the evaluation of performance can vary drastically depending on the objectives and domains of application. Working in an agricultural context, specifically on the problem of the automatic detection of plants in proximal sensing images, we studied twelve evaluation metrics that we used to evaluate three image segmentation models recently presented in the literature. After a unified presentation of these metrics, we carried out an exploratory analysis of their relationships using a correlation analysis, a clustering of variables, and two factorial analyses (namely principal component analysis and multiple factorial analysis). We distinguished three groups of highly linked metrics and, through visual inspection of the representative images of each group, identified the aspects of segmentation that each group evaluates. The aim of this exploratory analysis was to provide some clues to practitioners for understanding and choosing the metrics that are most relevant to their agricultural task.

Diagnostic validity of different gonial angle segmentation for the assessment of mandibular growth direction: A retrospective study

BackgroundThe development of the mandible largely determines the sagittal and vertical lateral appearance. The gonial angle (Articulare-Gonion-Menton, Ar-Go’-Me), as a composite indicator, represents the growth direction of the mandible. We proposed a method based on the Frankfort horizontal (FH) plane and its vertical plane (VFH) to divide the gonial angle into sagittal and vertical components (Articulare-Gonion-VFH / Menton-Gonion-FH, Ar-Go'-VFH / Me-Go'-FH) and to compare the accuracy of diagnosing the development of the mandible and maxillofacial structures with other methods. MethodsLateral cephalometric films from 736 volunteers aged 6–30 years were collected and analyzed for cephalometric measurements. Four groups of segmentation-based angle, including the FH-based segmentation (Ar-Go'-VFH / Me-Go'-FH), the SN-based segmentation (Articulare-Gonion- Sellion-Nasion plane’ vertical plane/ Menton-Gonion- Sellion-Nasion plane, Ar-Go'-VSN / Me-Go'-SN), the Go’-S based segmentation(Articulare-Gonion-Sellion / Menton-Gonion-Sellion, Ar-Go'-S / Me-Go'-S), and the Go’-N based segmentation (Articulare-Gonion-Nasion / Menton-Gonion-Nasion, Ar-Go'-N / Me-Go'-N), as well as commonly used sagittal and vertical indices were measured. Pearson correlation analysis was used to show the representativeness of different segmentation methods on the mandibular growth direction. ResultsAs the gonial angle decreased with age, all the segmentation-based metrics decreased. The plane-based segmentation metrics, including Ar-Go'-VFH / Me-Go'-FH, Ar-Go'-VSN / Me-Go'-SN, were superior to the point-based segmentation metrics (Ar-Go'-S / Me-Go'-S, and Ar-Go'-N / Me-Go '-N) in evaluating vertical and sagittal development of the mandible. The sagittal indicators displayed alteration of ramus and condyle, while these vertical indicators responded to the alteration of the mandibular corpus and gonial angle. ConclusionsThe gonial angle should be clinically segmented with planes (including SN plane and FH plane) rather than points (including Go'-S and Go'-N) to assess mandibular development. The FH plane-based segmentation method facilitated chair-side diagnosis of the mandibular growth direction.

A comprehensive review of computer-aided whole-slide image analysis: from datasets to feature extraction, segmentation, classification and detection approaches

With the development of Computer-aided Diagnosis (CAD) and image scanning techniques, Whole-slide Image (WSI) scanners are widely used in the field of pathological diagnosis. Therefore, WSI analysis has become the key to modern digital histopathology. Since 2004, WSI has been used widely in CAD. Since machine vision methods are usually based on semi-automatic or fully automatic computer algorithms, they are highly efficient and labor-saving. The combination of WSI and CAD technologies for segmentation, classification, and detection helps histopathologists to obtain more stable and quantitative results with minimum labor costs and improved diagnosis objectivity. This paper reviews the methods of WSI analysis based on machine learning. Firstly, the development status of WSI and CAD methods are introduced. Secondly, we discuss publicly available WSI datasets and evaluation metrics for segmentation, classification, and detection tasks. Then, the latest development of machine learning techniques in WSI segmentation, classification, and detection are reviewed. Finally, the existing methods are studied, and the application prospects of the methods in this field are forecasted.

Deep learning based high-throughput phenotyping of chalkiness in rice exposed to high night temperature

BackgroundRice is a major staple food crop for more than half the world’s population. As the global population is expected to reach 9.7 billion by 2050, increasing the production of high-quality rice is needed to meet the anticipated increased demand. However, global environmental changes, especially increasing temperatures, can affect grain yield and quality. Heat stress is one of the major causes of an increased proportion of chalkiness in rice, which compromises quality and reduces the market value. Researchers have identified 140 quantitative trait loci linked to chalkiness mapped across 12 chromosomes of the rice genome. However, the available genetic information acquired by employing advances in genetics has not been adequately exploited due to a lack of a reliable, rapid and high-throughput phenotyping tool to capture chalkiness. To derive extensive benefit from the genetic progress achieved, tools that facilitate high-throughput phenotyping of rice chalkiness are needed.ResultsWe use a fully automated approach based on convolutional neural networks (CNNs) and Gradient-weighted Class Activation Mapping (Grad-CAM) to detect chalkiness in rice grain images. Specifically, we train a CNN model to distinguish between chalky and non-chalky grains and subsequently use Grad-CAM to identify the area of a grain that is indicative of the chalky class. The area identified by the Grad-CAM approach takes the form of a smooth heatmap that can be used to quantify the degree of chalkiness. Experimental results on both polished and unpolished rice grains using standard instance classification and segmentation metrics have shown that Grad-CAM can accurately identify chalky grains and detect the chalkiness area.ConclusionsWe have successfully demonstrated the application of a Grad-CAM based tool to accurately capture high night temperature induced chalkiness in rice. The models trained will be made publicly available. They are easy-to-use, scalable and can be readily incorporated into ongoing rice breeding programs, without rice researchers requiring computer science or machine learning expertise.

Automated food intake tracking requires depth-refined semantic segmentation to rectify visual-volume discordance in long-term care homes

Malnutrition is a multidomain problem affecting 54% of older adults in long-term care (LTC). Monitoring nutritional intake in LTC is laborious and subjective, limiting clinical inference capabilities. Recent advances in automatic image-based food estimation have not yet been evaluated in LTC settings. Here, we describe a fully automatic imaging system for quantifying food intake. We propose a novel deep convolutional encoder-decoder food network with depth-refinement (EDFN-D) using an RGB-D camera for quantifying a plate’s remaining food volume relative to reference portions in whole and modified texture foods. We trained and validated the network on the pre-labelled UNIMIB2016 food dataset and tested on our two novel LTC-inspired plate datasets (689 plate images, 36 unique foods). EDFN-D performed comparably to depth-refined graph cut on IOU (0.879 vs. 0.887), with intake errors well below typical 50% (mean percent intake error: -4.2%). We identify how standard segmentation metrics are insufficient due to visual-volume discordance, and include volume disparity analysis to facilitate system trust. This system provides improved transparency, approximates human assessors with enhanced objectivity, accuracy, and precision while avoiding hefty semi-automatic method time requirements. This may help address short-comings currently limiting utility of automated early malnutrition detection in resource-constrained LTC and hospital settings.

Adaptive Denoising and Alignment Agents for Infrared Imaging

Clinical infrared imagers, such as those used to visualize subcutaneous vasculature, rely on the image-projection feedback loops. Specifically, these instruments embed model-based feedback algorithms to process the acquired ’invisible’ infrared data streams and then to project a visible copy of them back to the imaged region (e.g., the skin of the patient’s forearm). Being inherently noisy, the infrared frames and their projections are prone to misalignment, demanding frequent instrument tuning and recalibration. To address this challenge, we propose to reconsider the feedback loop entailed in such imagers from the standpoint of multi-agent deep learning. Both proposed agents – <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Denoiser</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Aligner</i> – are adaptive in that they continuously optimize the corresponding target value functions. Namely, the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Denoiser</i> learns the proper frequency decomposition of the acquired infrared data until the target segmentation metric is maximized; whereas, the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Aligner</i> learns the intensity fluctuations within the segmentation mask tuned by the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Denoiser</i> until its maximal overlap with the source infrared image. The idea is validated synthetically on a benchmark dataset and experimentally on a bench-top vein scanner in the lab, with the duet of agents proving efficient in handling the distortions and the misalignment.

Improved retinal vessel segmentation using the enhanced pre-processing method for high resolution fundus images

Background: By diagnosing using fundus images, ophthalmologists can possibly detect symptoms of retinal diseases such as diabetic retinopathy, age-related macular degeneration, and retinal detachment. A number of studies have also found some links between fundus image analysis data and other underlying systemic diseases such as cardiovascular diseases, including hypertension and kidney dysfunction. Now that imaging technology is advancing further, more fundus cameras are currently equipped with the capability to produce high resolution fundus images. One of the public databases for high-resolution fundus images called High-Resolution Fundus (HRF) is consistently used for validating vessel segmentation algorithms. However, it is noticed that the segmentation outputs from the HRF database normally include noisy pixels near the upper and lower edges of the image. In this study, we propose an enhanced method of pre-processing the images so that these noisy pixels can be eliminated, and thus the overall segmentation performance can be increased. Without eliminating the noisy pixels, the visual segmentation output shows a large number of false positive pixels near the top and bottom edges. Methods: The proposed method involves adding additional padding to the image before the segmentation procedure is applied. In this study, the Bar-Combination Of Shifted FIlter REsponses (B-COSFIRE) filter is used for retinal vessel segmentation. Results: Qualitative assessment of the segmentation results when using the proposed method showed improvement in terms of noisy pixel removal from near the edges. Quantitatively, the additional padding step improves all considered metrics for vessel segmentation, namely Sensitivity (73.76%), Specificity (97.53%), and Matthew’s Correlation Coefficient (MCC) value (71.57%) for the HRF database. Conclusions: Findings from this study indicate improvement in the overall segmentation performance when using the proposed double-padding method of pre-processing the fundus image prior to segmentation. In the future, more databases with various resolutions and modalities can be included for further validation.

Segmentation of Polyp Instruments using UNet based deep learning model

In this paper, we present a UNet architecture-based deep learning method that is used to segment polyp and instruments from the image data set provided in the MedAI Challenge2021. For the polyp segmentation task, we developed a UNet based algorithm for segmenting polyps in images taken from endoscopies. The main focus of this task is to achieve high segmentation metrics on the supplied test dataset. Similarly for the polyp segmentation task, in the instrument segmentation task, we have developed UNet based algorithms for segmenting instruments present in colonoscopy videos.

3D Attention M-net for Short-axis Left Ventricular Myocardium Segmentation in Mice MR cardiac Images.

Small rodent cardiac magnetic resonance imaging (MRI) plays an important role in preclinical models of cardiac disease. Accurate myocardial boundaries delineation is crucial to most morphological and functional analysis in rodent cardiac MRIs. However, rodent cardiac MRIs, due to animal's small cardiac volume and high heart rate, are usually acquired with sub-optimal resolution and low signal-to-noise ratio (SNR). These rodent cardiac MRIs can also suffer from signal loss due to the intra-voxel dephasing. These factors make automatic myocardial segmentation challenging. Manual contouring could be applied to label myocardial boundaries but it is usually laborious, time consuming, and not systematically objective. In this study, we present a deep learning approach based on 3D attention M-net to perform automatic segmentation of left ventricular myocardium. In the deep learning architecture, we use dual spatial-channel attention gates between encoder and decoder along with multi-scale feature fusion path after decoder. Attention gates enable networks to focus on relevant spatial information and channel features to improve segmentation performance. A distance derived loss term, besides general dice loss and binary cross entropy loss, was also introduced to our hybrid loss functions to refine segmentation contours. The proposed model outperforms other generic models, like U-Net and FCN, in major segmentation metrics including the dice score (0.9072), Jaccard index (0.8307) and Hausdorff distance (3.1754 pixels), which are comparable to the results achieved by state-of-the-art models on human cardiac ACDC17 datasets.Clinical relevance Small rodent cardiac MRI is routinely used to probe the effect of individual genes or groups of genes on the etiology of a large number of cardiovascular diseases. An automatic myocardium segmentation algorithm specifically designed for these data can enhance accuracy and reproducibility of cardiac structure and function analysis.

Encoder-Decoder Architecture for Ultrasound IMC Segmentation and cIMT Measurement.

Cardiovascular diseases (CVDs) have shown a huge impact on the number of deaths in the world. Thus, common carotid artery (CCA) segmentation and intima-media thickness (IMT) measurements have been significantly implemented to perform early diagnosis of CVDs by analyzing IMT features. Using computer vision algorithms on CCA images is not widely used for this type of diagnosis, due to the complexity and the lack of dataset to do it. The advancement of deep learning techniques has made accurate early diagnosis from images possible. In this paper, a deep-learning-based approach is proposed to apply semantic segmentation for intima-media complex (IMC) and to calculate the cIMT measurement. In order to overcome the lack of large-scale datasets, an encoder-decoder-based model is proposed using multi-image inputs that can help achieve good learning for the model using different features. The obtained results were evaluated using different image segmentation metrics which demonstrate the effectiveness of the proposed architecture. In addition, IMT thickness is computed, and the experiment showed that the proposed model is robust and fully automated compared to the state-of-the-art work.

Stratification by Tumor Grade Groups in a Holistic Evaluation of Machine Learning for Brain Tumor Segmentation.

Accurate and consistent segmentation plays an important role in the diagnosis, treatment planning, and monitoring of both High Grade Glioma (HGG), including Glioblastoma Multiforme (GBM), and Low Grade Glioma (LGG). Accuracy of segmentation can be affected by the imaging presentation of glioma, which greatly varies between the two tumor grade groups. In recent years, researchers have used Machine Learning (ML) to segment tumor rapidly and consistently, as compared to manual segmentation. However, existing ML validation relies heavily on computing summary statistics and rarely tests the generalizability of an algorithm on clinically heterogeneous data. In this work, our goal is to investigate how to holistically evaluate the performance of ML algorithms on a brain tumor segmentation task. We address the need for rigorous evaluation of ML algorithms and present four axes of model evaluation—diagnostic performance, model confidence, robustness, and data quality. We perform a comprehensive evaluation of a glioma segmentation ML algorithm by stratifying data by specific tumor grade groups (GBM and LGG) and evaluate these algorithms on each of the four axes. The main takeaways of our work are—(1) ML algorithms need to be evaluated on out-of-distribution data to assess generalizability, reflective of tumor heterogeneity. (2) Segmentation metrics alone are limited to evaluate the errors made by ML algorithms and their describe their consequences. (3) Adoption of tools in other domains such as robustness (adversarial attacks) and model uncertainty (prediction intervals) lead to a more comprehensive performance evaluation. Such a holistic evaluation framework could shed light on an algorithm's clinical utility and help it evolve into a more clinically valuable tool.

Sashimi: A toolkit for facilitating high‐throughput organismal image segmentation using deep learning

Abstract Digitized specimens are an indispensable resource for rapidly acquiring big datasets and typically must be pre‐processed prior to conducting analyses. One crucial image pre‐processing step in any image analysis workflow is image segmentation, or the ability to clearly contrast the foreground target from the background noise in an image. This procedure is typically done manually, creating a potential bottleneck for efforts to quantify biodiversity from image databases. Image segmentation meta‐algorithms using deep learning provide an opportunity to relax this bottleneck. However, the most accessible pre‐trained convolutional neural networks (CNNs) have been trained on a small fraction of biodiversity, thus limiting their utility. We trained a deep learning model to automatically segment target fish from images with both standardized and complex, noisy backgrounds. We then assessed the performance of our deep learning model using qualitative visual inspection and quantitative image segmentation metrics of pixel overlap between reference segmentation masks generated manually by experts and those automatically predicted by our model. Visual inspection revealed that our model segmented fishes with high precision and relatively few artifacts. These results suggest that the meta‐algorithm (Mask R‐CNN), in which our current fish segmentation model relies on, is well suited for generating high‐fidelity segmented specimen images across a variety of background contexts at rapid pace. We present Sashimi, a user‐friendly command line toolkit to facilitate rapid, automated high‐throughput image segmentation of digitized organisms. Sashimi is accessible to non‐programmers and does not require experience with deep learning to use. The flexibility of Mask R‐CNN allows users to generate a segmentation model for use on diverse animal and plant images using transfer learning with training datasets as small as a few hundred images. To help grow the taxonomic scope of images that can be recognized, Sashimi also includes a central database for sharing and distributing custom‐trained segmentation models of other unrepresented organisms. Lastly, Sashimi includes both auxiliary image pre‐processing functions useful for some popular downstream color pattern analysis workflows, as well as a simple script to aid users in qualitatively and quantitatively assessing segmentation model performance for complementary sets of automatically and manually segmented images.

Multicriteria Evaluation of Deep Neural Networks for Semantic Segmentation of Mammographies

Breast segmentation plays a vital role in the automatic analysis of mammograms. Accurate segmentation of the breast region increments the probability of a correct diagnostic and minimizes computational cost. Traditionally, model-based approaches dominated the landscape for breast segmentation, but recent studies seem to benefit from using robust deep learning models for this task. In this work, we present an extensive evaluation of deep learning architectures for semantic segmentation of mammograms, including segmentation metrics, memory requirements, and average inference time. We used several combinations of two-stage segmentation architectures composed of a feature extraction net (VGG16 and ResNet50) and a segmentation net (FCN-8, U-Net, and PSPNet). The training examples were taken from the mini Mammographic Image Analysis Society (MIAS) database. Experimental results using the mini-MIAS database show that the best net scored a Dice similarity coefficient of 99.37% for breast boundary segmentation and 95.45% for pectoral muscle segmentation.